HydraDB × Connectors Hackathon · Evidence-backed retrieval
QueueProof
Ask one plain-language question across every source you own and get an answer where each claim opens back to the record it came from.
The problem
Answers that cross tools lose their evidence. Ask a question that touches GitHub, Linear, Slack, Gmail and a folder of documents, and most tools hand you a confident paragraph with no way to check where any of it came from. When the sources disagree, the disagreement gets smoothed over instead of surfaced.
How it works
QueueProof retrieves through HydraDB's cross-source evidence layer and returns a concise answer whose every claim opens to the original source record behind it. Connectors are verified against attributable records rather than saved credentials, and evidence is merged without collapsing distinct entities into one.
A degraded connector or a contradiction between sources stays visible instead of being hidden. The same read contract compiles into a deterministic priority queue and is exposed through MCP so AI clients can consume it directly. Writes stay behind an explicit approval boundary and only count as executed once a provider response ID has been stored.
Highlights
- Claim-level citationsEvery sentence in the answer links to the exact record it was drawn from.
- Connector receiptsSources are verified against attributable records, not against stored credentials.
- Deterministic queueThe read contract compiles into a repeatable priority queue and an MCP surface.
- Approval-gated writesNothing is written until you approve it, and a write only counts once its response ID is stored.